Fingerprint matching in mobile devices (Android platform)

edit

One of the more important step in the fingerprint identification is acquisition of fingerprint. In the normal state we use live-scan for acquisition but you will acquire the fingerprint by mobile phone camera . The quality determination of these pictures are difficult. The quality determination fingerprint is done by analysis of ridge & valley frequency(DFT) & area(contourArea).

Note : In the general for all phone camera the preprocessing step is difficult . The most processes in the fingerprint field is block-wise for set a block(region of interest) on image you can use cvSetImageROI.

The following steps are required for fingerprint identification.

• normalize the input image by equalizeHist

a) input image b)normalized image

• Enhance the normalize image by block-wise gabor filter according to block-wise orientation map then binarize the fingerprint image.

a) input image b) enhanced image by gabor filter

Orientation map computes from block-wise PCA(principal component analysis) or block-wise gradient direction(cartToPolar.

• create skeleton map from enhanced binary image.

• Feature extraction from skeleton map .The feature in the fingerprint skeleton is Minutia.

Note : The regular minutias are ridge-end & bifurcation

• Determine the class of fingerprint by singular points from orientation map.Singular points are the regions where large changes of orientation happens & they are classified in core & delta.

singular points are useful in the clustering & alignment of fingerprint .A common method to estimate the singular points is poincare.

• search of fingerprint is done by position of minutia relative to each other .This search is rotation invariant. because of the high number of fingerprint & high number of features in each fingerprint for search you need use to kd-tree & hash(locality sensitive hash method).

For search operation,Create a connection between each minutia and other neighbor minutias by specefic distance. Then compute angles(a,b) & distance between two minutias l then create feature 3* mintia count and searh by above classifier.

Minutias j and k